Response time (RT) analysis in statement verification tasks is a promising framework for investigating mental models. Participants are asked to answer as quickly as possible if specially selected statements are true or false. Recent research demonstrates that the simultaneous activation of conflicting mental models results in a higher mistake rate and a longer RT. However, an important feature of mental models as persistent constructs was not emphasized. In order to study that, demonstrating multiple semantically related statements separated in time is preferable in comparison to single statements. The hypotheses of mental model persistence were tested in two experiments. Experiment 1 reveals that ΔRT for coherent responses to semantically related statements is significantly lower than for incoherent ones. This finding supports the hypothesis of the persistence of mental models. Our first experiment had certain procedural limitations. The random order in which the statements were presented resulted in a different number of other statements between related statements. To test the hypotheses about the mechanisms underlying the obtained effects and to reproduce the main effects after eliminating methodological flaws, Experiment 2 was carried out. In Experiment 2, the finding was replicated. Also, it was demonstrated that ΔRT for coherent answers was lower for both conflicting and congruent statements. Possible mechanisms of mental models functioning are discussed.